ext3, xfs, or reiserfs, 200GB or 400GB?

Man of Honour
Joined
4 Nov 2002
Posts
15,513
Location
West Berkshire
I'm getting ready to rebuild my RAID array with upgraded disks (400GB instead of 200GB).

Note - there are no operating system files on the RAID array - it's for storage only.

I've used both reiserfs and ext3 for years, and had no issues with either, however, since I get to make the choice again, are there any particular reason(s) why I would pick one over the other - performance, reliability, ease of fixing if something breaks, or anything else? I'm aware of the features of each, but just want to check others' experience in case I just got lucky in some way. :)

Also, I'm undecided as to whether to go with 2x 200GB partitions or just one big 400GB. In the past (including the old 200GB disks), I've split disks into smaller partitions, but that's mainly habit (there were times when you had no choice due to operating system limits). With more recent installs, I've split only for security, not size. So, are there any performance/reliability benefits or risks from going either way?
 
It's generally accepted that if you're using big files and have a ups, you want XFS. If you have tiny files and have a UPS, Reiser4 is the way to go. Without a UPS, you probably want Ext3 as that recovers best in the instance of a power failure.

People tend to state the performance of reiser4 as being in the region of "amazing" in general, but others claim it has stability issues. I suppose if stability is your concern, you're usually going to win with ext3. If you want performance at the possible cost of stability, something more exotic is better.

Generally, if you have the cash, everything important should be backed up on secure (not-fast) media (tape, whatever), and your working copy should be high-performance. Best of both worlds then ;)
 
I do have a UPS on this system (unusal for a home system, but there), and I have both lots of very small and lots of very big files (I have several hundred files over 100MB each, and I have many thousand files under 32K each).

Reiser4 is certainly interesting, but is it in the kernel yet? I have a 'golden rule' of never modifying kernel sources for any reason - including adding drivers. It just causes pain when you later decide to update.

I do want performance, but not at the expensive of reliability. I am now (and still will) use RAID 1 for data security, so risking a potentially flaky filesystem would defeat the point. I'm not the most efficient at making backups, so having a bit of redundancy instead seems like a good idea (and has saved my bacon at least once before).
 
Hi,

at work on on small FC SANS less than 1TB storage, over 12 disks have found both Ext3 and XFS to be good it's really swings and round abouts as the saying goes, overall i have personally found XFS to be better than EXt3 as i have found it quicker in read / writes and also in recovering after after pulling the plug. Could always do your own test?

Cheers
Deano

[edit] splitting the disks, we use multiple raid 5 arrays with lvm, which is probably OTT, unless you want to add more storage later. probably go with 1 400gb partion unless using lvm and two 200gb disks. HTH.
 
Last edited:
Looks like the article above agrees with XFS. I've never come across XFS before. Looks like it might be a good choice. :)
 
deano said:
[edit] splitting the disks, we use multiple raid 5 arrays with lvm, which is probably OTT, unless you want to add more storage later. probably go with 1 400gb partion unless using lvm and two 200gb disks. HTH.
Yeah, lvm is probably pushing it a bit, though I have used it. That being the case, my choice is either 2x 400GB or 4x 200GB. :)
 
Adaptec 2400A. Proper hardware RAID. ;)

I'm going RAID 1 as that leaves me in good shape if the controller ever gives up (RAID 5 is better for disk capacity, but very much relies on the controller). It'll also prove handy when I do the swap as I'll just swap out all four disks, and put one off each channel in a USB caddy to get the data copied off. :)
 
Berserker said:
Yeah, lvm is probably pushing it a bit, though I have used it. That being the case, my choice is either 2x 400GB or 4x 200GB. :)

I'm no expert but have always been under the belief that ext3 will give the best reliability so that's all I ever use.

As for LVM, definitely use it. It will make your life so much easier when you need to resize a partition or add more capacity. Best thing I did to my file server was to start using LVM!
 
if you have not already got the 2400A then ahve a look at the 3ware cards as they are very good!!

Are you going to be running it in a 100/133mhz pci slot? if your just running it on standard pci it won't take much to flood your bus, then your start getting bottle necks.

I'd go for the 4 x 200gb disks using Xfs and LVM.

cheers
Deano
 
deano said:
if you have not already got the 2400A then ahve a look at the 3ware cards as they are very good!!

Are you going to be running it in a 100/133mhz pci slot? if your just running it on standard pci it won't take much to flood your bus, then your start getting bottle necks.

I'd go for the 4 x 200gb disks using Xfs and LVM.

cheers
Deano
Yeah, 3ware are good, but I've had the 2400A for 30 months, and it too is good.

Oh, and yes, I've noticed the PCI bus getting a bit knackered. Only using cheapo mobo though, so not surprising.

I'm surprised at the move away from reiserfs. A few years ago it was the done thing. Now it seems xfs has taken that spot.

No objection to using lvm, but I don't see the benefit in doing so in this instance (given that I'll be fully populating the RAID).
 
I always thought Reiserfs was good for lots of small files, like mp3 collections? Can't remember where I read that but probably wrong as ext3 seems to be default in a lot of Linux distros these days.

I'm using IDE Raid-0 from my MSI K8N Neo2 Platinum motherboard so looking into dual booting Windows XP and some form of Linux, probably from Novell due to work requirements. Will see how I get on later today for trying it out.
 
LVM probably not requird by the sounds of things then, unless you want to use lvm snapshot as a back up tool unless you got a spare lto3 drive :)

if you are using the data over the network may be worth your while investing in a newer board, as built on network cards sit on the pci bus.

supermicro AMD boards using the server works HT2000 / HT1000 chipsets are good or go for one with a pro nvidia chipset.

have found XFS to be good with file sizes ranging from 100k to 2-3gb in size, seems less delay than with ext3 :)

cheers
deano
 
I've used XFS on my 0.5 TB server for years. It's mature, well supported, very fast, directory listings are quick (I have some folders with 10's of thousands of files). Moving gigs of files is instant. (Using NTFS at work on a 1TB 10 (or 12?) disk SCSI array takes a hell of a lot longer to move files about and display a listing)

In short it rocks. I am currently running a partition of XFS and a partition of resier3 (thought id's experiment) and tbh I can't really see that much difference, but i'll stick with xfs until reiser4 is fully supported and tested.
 
afraser2k said:
I always thought Reiserfs was good for lots of small files, like mp3 collections?
Reiser4 rocks for lots of small files - we run it on our World Wind image server which contains millions of tiny images :)
 
Back
Top Bottom